AI won’t replace clinical formulation, but it may help clinicians see their thinking more clearly
Dr Chris Ludlow on turning session transcripts into causal diagrams, and why the underlying model matters as much as the transcript
Lecturer in Psychology & Deputy Director, Swinburne Psychology Clinic, Swinburne University of Technology
For psychologists, provisional psychologists and clinical supervisors who write formulations as a matter of course, a familiar trade-off shows up on a busy day: the narrative note wins by default, because nobody has time to draw a diagram at 4pm on a Thursday. Dr Chris Ludlow, Lecturer in Psychology at Swinburne University of Technology and Deputy Director of the Swinburne Psychology Clinic, argues that trade-off is worth resisting, and that AI can help clinicians resist it without doing their thinking for them.
Diagrams are cognitive tools, not just communication aids
Engineers do not build bridges from paragraphs of prose. They use diagrams to make relationships, risks and decisions visible before a single beam is poured. Psychologists already lean on the same instinct: CBT maintenance cycles, DBT chain analyses, family therapy genograms and the reciprocal-role diagrams used in cognitive analytic therapy are all attempts to externalise a case rather than describe it in running text.
That externalising step is not decorative. Cognitive load theory holds that working memory is a narrow bottleneck, and that offloading part of a reasoning task onto an external representation frees capacity for the harder work of comparing, revising and critiquing (Kirschner, Sweller, Kirschner, & Zambrano, 2018).
The webinar Enhancing Clinical Reasoning Using AI and Formulation Diagrams walks through this reasoning live, contrasts it with narrative-only formulation, and demonstrates Just Ask NovoNote generating diagrams from a transcript in real time.
The practical limitation: sustaining diagrams across a caseload
Engineering diagrams get extended planning periods and comparatively stable specifications. Clinical diagrams do not. Clinicians hold multiple open cases at once, each formulation needs revising as new information comes in, and manually redrawing a diagram for every session is not something most services can sustain. That is the practical reason narrative notes still win by default, not a lack of appetite for diagrams.
From transcript to diagram: what changes with AI
What AI-assisted tools such as Just Ask NovoNote change is not the value of a diagram, but the cost of producing one. Feed in a session transcript, and the system proposes a formulation diagram directly from the clinical content, without a clinician manually redrawing boxes and arrows.
An illustration from the 1965 training film Three Approaches to Psychotherapy (Shostrom, 1965) makes the point cleanly. In the session, Carl Rogers interviews “Gloria” about her discomfort concealing a new relationship from her daughter. Run through a causal-formulation prompt, the transcript resolves into a trigger (thoughts about her daughter discovering the relationship), an underlying belief (“I need external approval to be acceptable”), a chain of automatic thoughts, the resulting emotions (guilt, anxiety, shame), the behaviours those emotions drive (avoidance and deception), and the consequences that follow (increased distance, unresolved conflict). None of that is new content. It is the same session a clinician would formulate by hand. What changes is how quickly the structure becomes visible.
Comparing lenses, not locking in one
A second, less obvious use is comparison. The same transcript can be run through multiple therapeutic models rather than one: a CBT maintenance cycle, an interpersonal-therapy formulation built around role transition, a psychodynamic account of transference and defence, a solution-focused version built around exceptions and scaling questions, an internal-family-systems map of parts and self, and a cognitive-analytic reciprocal-role pattern, all generated from the same session transcript. Clinicians rarely have time to draft six competing formulations of one case by hand. AI-generated diagrams make that comparison practical, which matters because different models surface different treatment targets from the same material.
Where the diagrams earn their place clinically
The clearest use case is differential formulation: taking one presenting picture and laying two plausible pathways side by side. A child who refuses to get dressed, insists on routine, argues with adults and blames others could be formulated along an autism-spectrum pathway (sensory sensitivity, cognitive rigidity, difficulty reading others’ perspectives) or an oppositional-defiance pathway (a negative-reinforcement cycle, a stance against authority, difficulty that is situational rather than lifelong). Seeing both formulations side by side, with their distinguishing features made explicit, is a different task from holding both possibilities in mind while reading a case file.
The same underlying move, turning a tangle of information into a structure that can be checked against the evidence, applies to genograms in multisystemic family work and to separating current medical context from developmental history in an educational or neuropsychological assessment.
Cognitive offloading, not cognitive outsourcing
The distinction that matters most is between a tool that supports reasoning and one that replaces it. Handwriting a calculation on paper is cognitive offloading: the tool holds information so working memory does not have to, while the person keeps doing the reasoning. Following a turn-by-turn GPS direction sits closer to cognitive outsourcing: the tool does the spatial reasoning, and the relevant skill measurably disengages the less it gets used (Kirschner et al., 2018).
Formulation diagrams sit closer to the first category, provided clinicians treat them that way. An AI-generated diagram proposes a structure; it does not verify that the structure is clinically correct. Clinicians in this space keep returning to the same conclusion: AI can rapidly transform a transcript into a diagram and make reasoning visible and easier to communicate, but the clinician remains responsible for interpretation, judgement and determining clinical fit. A diagram that goes unchallenged is not a formulation. It is a plausible-looking output nobody has checked.
Formulation as a thinking tool
Clinical formulations are not only communication tools, they are thinking tools first. Treating them that way changes what a diagram is for: not a tidy visual to hand a client at the end of a session, but scaffolding built during the reasoning itself, open to revision the moment new information contradicts it.
What AI changes is not the reasoning itself. It changes how much of the mechanical work of building and rebuilding that scaffolding a clinician has to do by hand, which frees attention for the part of the job that was never going to be automated: deciding which formulation actually fits the person in front of you.
"AI suggests hypotheses. Clinicians decide what fits."
References
Henggeler, S. W., Schoenwald, S. K., Borduin, C. M., Rowland, M. D., & Cunningham, P. B. (2009). Multisystemic therapy for antisocial behavior in children and adolescents (2nd ed.). Guilford Press.
Kirschner, P. A., Sweller, J., Kirschner, F., & Zambrano R., J. (2018). From cognitive load theory to collaborative cognitive load theory. International Journal of Computer-Supported Collaborative Learning, 13(2), 213–233. https://doi.org/10.1007/s11412-018-9277-y
Ludlow, C. (2025). Investigating the capability of large language models to identify causal relations in psychiatric case studies: A methodological proof of concept for the analysis of psychological case formulations. PsyArXiv. https://doi.org/10.31234/osf.io/wfmv8_v1
Shostrom, E. L. (Director). (1965). Three approaches to psychotherapy [Film]. Psychological Films.
